It's often said that there are only two nonabelian groups of order 8 up to isomorphism, one is the quaternion group, the other given by the relations $a^4=1$, $b^2=1$ and $bab^{-1}=a^3$.

I've never understood why these are the only two. Is there a reference or proof walkthrough on how to show any nonabelian group of order 8 is isomorphic to one of these?

  • 8
    $\begingroup$ Since $G$ is a $p$-group, then $Z(G)$ must be nontrivial, and $G/Z(G)$ cannot be cyclic since $G$ is nonabelian. This should get you started. $\endgroup$ – Ted Sep 14 '11 at 7:30

A very down-to-Earth approach might be:

Let $G$ be a group of order 8.

Exercise 1: Show that the maximal order $m$ of an element $x$ of $G$ is either 2, 4, or 8.

Exercise 2: Show that if $m=2$, then the group $G$ is abelian.

Exercise 3: Show that if $m=8$, then the group $G$ is abelian.

Ok, so that leaves us with the case $m=4$. Let $x$ be an element of order 4. Let $H\simeq C_4$ be the subgroup generated by $x$.

Exercise 4: Show that $H$ is a normal subgroup of $G$.

Let $y\in G, y\notin H$ be a fixed element.

Exercise 5: Show that $y^2\in H$.

Exercise 6: Show that $yxy^{-1}$ is an element of order 4 in $H$.

Exercise 7: Show that either $yxy^{-1}=x$ or $yxy^{-1}=x^3$.

Exercise 8: Show that if $yxy^{-1}=x$, then $G$ is abelian.

Ok, so we must have $yxy^{-1}=x^3.$ Assume that this is the case in what follows.

Exercise 9: Show that if $y$ is of order 2, then $G$ is isomorphic to a dihedral group.

Exercise 10: Show that if $y$ is of order 4, then $y^2=x^2$.

Exercise 11: Show that if $y$ is of order 4, then $G$ is isomorphic to the quaternion group (or more precisely: the group of units of the Lipschitz' order)


Remark: You won't need the result of Exercise 5 until the two last ones. I just added it there, because the element $y$ was introduced at that point.

  • 4
    $\begingroup$ All the exercises are in some sense one-liners. Well, in 9 & 11 it is a matter of taste at which point you will have satisfied yourself. $\endgroup$ – Jyrki Lahtonen Sep 15 '11 at 12:25
  • $\begingroup$ I am glad to see your answer, because I was about to ask "how does one classify groups of order $8$ in an elegant way" on this site, found your answer and then notice I had the exact same proof written down but thought it was too long to be elegant... so perhaps this is the most elegant way to do it. ^_^ $\endgroup$ – Patrick Da Silva Mar 23 '12 at 21:21
  • $\begingroup$ @jyrki lahtoten I worked through all, but I have trouble with 5, any ideas? $\endgroup$ – user35687 Feb 12 at 20:52
  • 1
    $\begingroup$ @user35687 What do you know about quotient groups? Here $G/H$. Are you familiar with them at all. $\endgroup$ – Jyrki Lahtonen Feb 13 at 8:26

See www.math.uconn.edu/~kconrad/blurbs/grouptheory/groupsp3.pdf, which discusses groups of order p^3 for any prime p and treats the case p = 2 first.

  • $\begingroup$ I down-voted this post because I believe that the previous answer was more than sufficient. Although the general case is interesting, I think taking the time to outline a proof of it would have been more helpful than just providing a link to a full solution, especially as it included the $p=2$ case. $\endgroup$ – user1729 Sep 14 '11 at 13:54
  • 5
    $\begingroup$ @Swalbr: Why downvote a correct answer? If there is a good answer, this doesn't mean that there isn't another possible answer. This one is better, since is a generalization of the previous answer. And sometimes, when the question can be easily answered using a good google search a link to a complete written proof is better and time-saving for both parts. When someone asks a question it means that he/she wants to find an answer, and is willing to do some work about this. Some users here have the good habit of presenting only hints for solving a question, leaving some of the work for the OP. $\endgroup$ – Beni Bogosel Sep 15 '11 at 11:26
  • $\begingroup$ +1 from me, too. @Swlabr: It often happens in this site that after a correct answer somebody else shows up, and explains a more general case. Nothing wrong with that! The reason: it is difficult to estimate the level that the OP appreciates. There is so much variation worldwide that even targeting, say, freshman-level it is not clear exactly what is best. Also some other passers-by may, or I should say will, appreciate a different approach. I gambled on the level of first-course in algebraic structures, and built it around hints. The crowd seems to like it, but no final nod from OP, yet. $\endgroup$ – Jyrki Lahtonen Sep 15 '11 at 12:32
  • 1
    $\begingroup$ Please see meta.math.stackexchange.com/questions/415/…. I feel that although this is not a homework question there is no reason that it should not be treated as one. Indeed, I believe I was asked it as a homework question during my undergrad. $\endgroup$ – user1729 Sep 15 '11 at 14:05

Here's the proof that there are exactly five nonisomorphic groups of order $p^3$ for every prime $p$, as it appears in Marshall Hall's Theory of Groups.

  1. The abelian case is easy: you have $C_{p^3}$, $C_{p^2}\times C_p$, and $C_{p}\times C_p\times C_p$.

  2. The nonabelian case. There can be no element of order $p^3$, because then the group is cyclic.

    • If all elements are of order $p$, then $p$ must be odd (otherwise the group is abelian). The center of $G$ is of order $p$ (since the quotient must be of order $p^2$ and isomorphic to $C_p\times C_p$); let $x$ and $y$ be elements of $G$ whose images generate the two cyclic factors of $G/Z(G)$. Then $x^p = y^p = 1$, and $x^{-1}y^{-1}xy\neq 1$ (otherwise, $G$ would be abelian), but must be central; so $z=x^{-1}y^{-1}xy$ generates $Z(G)$. So $G$ is given by $$G = \langle x,y,z\mid x^p=y^p=z^p=1, xy=yxz,\ xz=zx,\ yz=zy\rangle.$$

    • If there is an element $x$ of order $p^2$, then $\langle x\rangle$ is a maximal abelian subgroup of $G$, and normal (since its index is the smallest prime that divides $|G|$. Let $b\notin A$. Then $b^p\in A$, and $b^{-1}ab=a^r$ for some $r$; since $G$ is nonabelian, $r\neq 1$. Since $b^p$ commutes with $a$, $a^{r^p}=a$, so $r^p\equiv 1\pmod{p^2}$. From Fermat's Little Theorem, $r^p\equiv r\pmod{p}$, so $r\equiv 1\pmod{p}$.

      Write $r=1+sp$, and let $j$ such that $js\equiv 1\pmod{p}$. Then $$b^{-j}ab^j = a^{r^j} = a^{(1+sp)^j} = a^{1+spj} =a^{1+p}.$$ Since $(j,p)=1$, $b^j\notin A$, so replacing $b$ with $b^j$, we may assume that $b^{-1}ab=a^{1+p}$.

      Now, $b^p = a^t$, and $t$ must be a multiple of $t$, because $b$ is not of order $p^3$. Write $t=up$, so $b^p=a^{up}$. Then we have: $$\begin{align*} (ba^{-u})^p &= b^pa^{-u(1+(1+p)+(1+p)^2 + \cdots + (1+p)^{p-1})}\\ &= b^p a^{-up-up(1+2+\cdots + p-1)}\\ &= b^p a^{-up-up\binom{p}{2}}. \end{align*}$$

      • If $p$ is odd, then $up\binom{p}{2}$ is a multiple of $p^2$, so we get $(ba^{-u})^p = b^pa^{-up} = b^pb^{-p} = 1$. Setting $c=ba^{-u}$ we get $c^{-1}ac = b^{-1}ab$, so the group is presented by $$\langle a,c\mid a^{p^2} = c^p = 1,\ ac = ca^{1+p}.\rangle$$

      • If $p$ is $2$, however, we get $(ba^{-u})^2 = b^2a^{-up-up} = b^2$. We have two possibilities: it could be that $b^2=1$, in which case we get the same presentation as above: $$\langle a,b\mid a^{4} = b^2 = 1,\ ab=ba^3\rangle.$$ Or it could be that $b^2=a^2$; we must have $b^{-1}ab=a^3$ (it cannot equal $a$, because then $a$ and $b$ commute and $G$ is abelian), so the group is given by $$\langle a,b\mid a^4=1,\ a^2=b^2,\ ab=ba^3.$$

Burnside uses essentially the same approach, though he only deals explicitly with odd $p$; the classification for groups of order $p^3$ takes up about two pages (one paragraph, but he invokes results covering two previous pages). He then proceeds to those of order $p^4$; that takes four and a half pages (plus invoking stuff that covers at least one previous page). He only lists those of order $2^3$ and $2^4$.


You probably meant $bab = a^3$ in the above.

The following steps should also work. In order not to spoil the fun for you, I omit all the details.

Let $G$ be a nonabelian group of order $8$ and let $Z$ be its centre.

(1) From the class equation we know that $|Z|$ is divisible by $2$, and hence $|Z|\gt1$

(2) If $G/Z$ were cyclic then $G$ would be abelian. Therefore only the case $|Z|=2$ is possible and we know that $G/Z$ must be isomorphic to the Klein group.

Now let $Z=\{1,z\}$ and take $a,b,c\in G$ with $G/Z = \{Z, aZ, bZ, cZ\}$. All squares a^2, b^2, c^2 lie in $Z$ and we may take $c=ab$. Now the rest is a case by case analysis

(3) We cannot have $a^2 = b^2 = (ab)^2 = 1$, because then every element of $G$ would have order $2$, forcing $G$ to be abelian. So at least one of those squares must equal $z$. In particular $G$ has an element of order $4$.

(4) If two of the above squares equal $z$, then so does the third. For instance, suppose $a^2 = z = b^2$. Then $(ab)^2 = 1$ would give $ab=zbaz=ba$ and $G$ would be abelian.

So now we arrived at two possibilities (up to permutation of $a$, $b$ and $c$):

case (i): $a^2 = b^2 = c^2 = z$. Then $G$ is the quaternion group.

case (ii) $a^2 = z$, $b^2 = 1$. Then $bab$ lies in $aZ$. Because of (4), $bab=a$ is impossible and therefore $bab = az = a^3$.


Let $G$ be non-abelian group of order 8.

By Sylow theorem, $G$ will have subgroups of order 2.

  • If $G$ has unique subgroup of order 2, then $G$ is quaternion (Ref. Theorem 12.5.2, Theory of groups- Marshall Hall). ($G$ can be cyclic also, but we have assumed it is non-abelian.)

  • If $G$ has more than one subgroups of order 2, then they will be $1+2k$ in number for some $k\geq 1$; so it is atleast $3$; let $H_1, H_2, H_3$ be three distinct subgroups of order $2$.

If $H_1\triangleleft G$, then $G/H_1$ is abelian (since its order is 4), we conclude that $[G,G]\leq H_1$, hence $H_1=[G,G]$ since $G$ is non-abelian.

Then observe that $H_2$ can-not be normal in $G$, otherwise $H_2=[G,G]=H_1$, contradiction.

(So we have found a subgroup of order $2$ which is not normal in $G$.)

$H_2$ is a subgroup of $G$ index 4, there is a homomorphism $\phi \colon G\rightarrow S_4$, with $ker(\phi)\subseteq H_2$ (See "Generalised Cayley's theorem" - Introduction to Theory of Groups, Rotman).

But $ker(\phi)\neq H_2$ since $H_2$ is not normal, so $ker(\phi)=\{1\}$; so $G$ is isomorphic to a subgroup of $S_4$ of order $8$; it is a Sylow-2 subgroup of $S_4$; and it is $D_8$ since when considering a square with vertices labelled $"1,2,3,4"$, its symmetries in terms of permutations give $D_8\leq S_4$.


$G$ is either quaternion or isomorphic to Sylow-2 subgroup of $S_4$ which is dihedral group of order 8.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.